As Anthropic prepares a record $100-billion initial public offering (IPO), Murthy Grandhi, company profiles analyst at GlobalData, offers this analysis.
Anthropic’s planned IPO could be one of the most unusual public listings yet. As a public benefit corporation (PBC), it must balance shareholder returns with its mission to develop advanced AI responsibly.
An independent body can elect a majority of the board, while seven co-founders hold Class F shares carrying 50,1% of votes on general shareholder matters. Public investors would therefore have limited influence over decisions that prioritise AI safety over commercial returns.
Growth meets intensifying competition
Anthropic generates revenue through developer APIs and products such as Claude and Claude Code, largely charging customers according to tokens processed. Revenue rose from $41 million in 2023 to approximately $16.3 billion in the first half of 2026. Usage-based billing accounted for 80% of first-half revenue, while 72% of revenue was billed in the US.
Models released in the previous six months generated more than 80% of September-quarter revenue. Claude Code is driving spending as AI agents tackle longer, more complex tasks. Nearly 12 000 customers spend over $100 000 annually, up from around 950 a year earlier, accounting for 74% of first-half revenue.
However, revenue growth slowed to about 50% in the September quarter after more than doubling in each of the previous two quarters. Although no customer accounted for over 10% of revenue in the first nine months of 2026, 48% flowed through Amazon, Google and Microsoft cloud marketplaces. Many customers lack long-term contracts and can reduce usage as competing models improve. Anthropic also estimates that customers violating its policies generated 5–7% of revenue between January and September before being cut off in August.
Improving economics, enormous commitments
Anthropic spent $8,9-billion on compute in the first half of 2026, including $2,4-billion supporting paying customers. It retained approximately 85 cents per revenue dollar after customer-serving compute costs, but just 28 cents after research and other operating expenses. Overall, it lost around four cents per revenue dollar. Revenue per dollar of compute spending improved from 40 cents in the March 2025 quarter to $2.08 in April–June 2026.
Reported net losses reached $80,7-billion in the first half of 2026, partly reflecting accounting adjustments on Amazon-held convertible notes valued at $108,6-billion at June-end. Despite holding $132-billion in cash and short-term investments, Anthropic faced nearly $266-billion in fixed or minimum compute commitments and approximately $321-billion in signed leases yet to commence.
An agreement involving SpaceX’s AI arm could eventually cost $3-billion monthly at full delivery. Amazon is simultaneously an investor, lender, cloud provider, marketplace operator, customer and competitor. Anthropic has committed to spend at least $110-billion on Amazon’s cloud.
The investment test
Anthropic must sustain growth as AI becomes cheaper and competition intensifies. Open-weight models are narrowing the gap with leading closed models, increasing pressure on pricing. Anthropic is expanding into life sciences, healthcare, finance and law, including through its acquisition of Coefficient Bio.
Investors must weigh exceptional demand and improving unit economics against concentrated control, enormous fixed commitments and commoditisation risks. Its public debut will test whether markets will embrace mission-first capitalism at unprecedented scale.